U.S. patent application number 11/088041 was filed with the patent office on 2005-09-22 for method and device for safeguarding a hazardous area.
Invention is credited to Dottling, Dietmar, Kruger, Lars, Progscha, Werner, Wendler, Martin, Wohler, Christian.
Application Number | 20050207618 11/088041 |
Document ID | / |
Family ID | 34986318 |
Filed Date | 2005-09-22 |
United States Patent
Application |
20050207618 |
Kind Code |
A1 |
Wohler, Christian ; et
al. |
September 22, 2005 |
Method and device for safeguarding a hazardous area
Abstract
In order to safeguard a hazardous area, in particular the
hazardous area of an automatically operating machine, images of the
hazardous area are recorded with at least a first and a second
image recording unit that are arranged offset from one another. The
images of the scene are subjected to three-dimensional scene
analyses using different three-dimensional scene analysis methods,
and the hazardous area is safeguarded when at least one of the
scene analysis methods detects a foreign object.
Inventors: |
Wohler, Christian;
(Heroldstatt, DE) ; Progscha, Werner; (Herrenberg,
DE) ; Kruger, Lars; (Ulm, DE) ; Dottling,
Dietmar; (Leinfelden-Echterdingen, DE) ; Wendler,
Martin; (Stuttgart, DE) |
Correspondence
Address: |
HARNESS, DICKEY & PIERCE, P.L.C.
P.O. BOX 828
BLOOMFIELD HILLS
MI
48303
US
|
Family ID: |
34986318 |
Appl. No.: |
11/088041 |
Filed: |
March 23, 2005 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
11088041 |
Mar 23, 2005 |
|
|
|
PCT/EP03/10279 |
Sep 16, 2003 |
|
|
|
Current U.S.
Class: |
382/103 |
Current CPC
Class: |
B25J 19/06 20130101;
B25J 9/1697 20130101; G06T 7/001 20130101; G06T 2207/30196
20130101; G06T 2207/30108 20130101; F16P 3/142 20130101; G06T
2207/10012 20130101; B25J 9/1674 20130101 |
Class at
Publication: |
382/103 |
International
Class: |
G06K 009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Sep 24, 2002 |
DE |
102 45 720.4 |
Apr 1, 2003 |
DE |
103 15 216.4 |
Claims
What is claimed is:
1. A method for safeguarding a hazardous area of an automatically
operating machine against intrusion of foreign objects, comprising
the steps of: providing at least a first and a second image
recording unit which are arranged offset from one another,
recording at least a first image of the hazardous area by means of
the first image recording unit and a second image of the hazardous
area by means of the second image recording unit, and carrying out
a first and a second scene analysis on the basis of the first and
second images in order to detect foreign objects in the hazardous
area, wherein the first and second scene analyses are carried out
using at least two algorithmically different methods, and wherein
the machine is brought into a safe state when at least one of the
scene analysis methods detects a foreign object.
2. The method of claim 1, wherein each of the scene analysis
methods is carried out using the first and the second images.
3. The method of claim 1, wherein one of the scene analysis methods
comprises a correlation-based scene analysis, and another one
comprises a contour-based scene analysis.
4. The method of claim 1, further comprising the step of providing
a reference image of the hazardous area, the reference image being
free from foreign objects to be detected, wherein at least one of
the scene analysis methods processes only such image regions in the
first and second images that differ significantly from the
reference image.
5. The method of claim 1, wherein at least three images of the
hazardous area are recorded using at least three image recording
units that are not arranged in line.
6. The method of claim 5, wherein the at least three images are
grouped into at least two image pairs, with each two image pairs
having one of the at least three images in common.
7. The method of claim 1, wherein all scene analysis methods are
applied to all the images.
8. The method of claim 1, further comprising the step of evaluating
a number of temporally successive groups of the at least first and
second images in order to determine at least one movement parameter
of a foreign object detected.
9. The method of claim 1, wherein an exposure time for each image
recording unit is set as a function of the image recorded by
it.
10. The method of claim 1, wherein a three-dimensional
representation of the scene is established by minimizing a target
function to be determined from all the images simultaneously.
11. A method for safeguarding a hazardous area, wherein at least
two images of the hazardous area are recorded by means of at least
a first and a second image recording unit, the image recording
units being arranged offset from one another, and wherein foreign
objects in the hazardous area are detected by means of a scene
analysis of the first and second images, wherein the scene analysis
is carried out on the basis of the at least two images using at
least two algorithmically different methods, and wherein the
hazardous area is safeguarded when at least one of the methods
detects a foreign object.
12. A device for safeguarding a hazardous area of an automatically
operating machine, having at least a first and a second image
recording unit for recording at least a first and a second image of
the hazardous area, the image recording units being arranged offset
from one another, having an evaluation unit adapted for detecting
foreign objects in the hazardous area by means of a scene analysis
on the basis of the first and second images, and having a switch
that is responsive to the evaluation unit in order to safeguard the
machine upon detection of a foreign object in the hazardous area,
wherein the evaluation unit is designed for carrying out the scene
analysis using at least two algorithmically different methods, and
wherein the switch is activated when at least one of the scene
analysis methods detects a foreign object.
13. The device of claim 12, wherein the evaluation unit is set up
to apply all the scene analysis methods to all of the images of the
scene.
14. The device of claim 12, further comprising at least a third
image recording unit for recording a third image of the hazardous
area, the evaluation unit being designed to apply at least one of
the two scene analysis methods to the at least three images.
15. The device of claim 12, wherein at least two of the image
recording units are arranged in a common housing.
16. The device of claim 12, wherein the evaluation unit is designed
for establishing a three-dimensional. representation of the scene
by minimizing a target function to be determined from all images
simultaneously.
Description
CROSSREFERENCES TO RELATED APPLICATIONS
[0001] This application is a continuation application to
international patent application PCT/EP2003/010279 filed on Sep.
16, 2003 designating the U.S., which international patent
application was published in German language and which claims
priority from German patent applications DE 102 45 720.4 filed on
Sep. 24, 2002 and DE 103 15 216.4 filed on Apr. 1, 2003.
BACKGROUND OF THE INVENTION
[0002] The present invention relates to a method and a device for
safeguarding a hazardous area, in particular the hazardous area of
an automatically operating machine. In more detail, the invention
relates to such a method and such a device using at least a first
and a second image recording unit for recording at least two images
of the hazardous area.
[0003] Hazardous areas, in particular hazardous areas of machines
and industrial plants, are usually safeguarded so far by using
light barriers, light curtains or laser scanners, often in
combination with additional mechanical barriers. However, such
protective measures require complicated installation and
adjustment. In addition, these protective measures are not very
flexible when it is desired to adapt the safeguarding of the
hazardous area to different operational situations.
[0004] In order to overcome these disadvantages, attempts have been
made for some time to safeguard a hazardous area by means of an
image recording unit, typically a digital camera, and by means of
intelligent image evaluation methods. Reference is made in this
regard to DE 199 38 639 A1, by way of example. This reference
discloses a device for safeguarding a hazardous area wherein a
virtual barrier is generated in front of the hazardous area by
means of the image recording unit and by means of an artificial,
"cooperative" target. Breaking through the barrier can be detected
with the image recording unit in the same way as in the case of a
light barrier. If appropriate, the machine from which the hazard
originates is switched off thereupon or put into a non-hazardous
state in some other way.
[0005] Although this known device is already substantially easier
to install and exhibits a higher level of flexibility than
conventional light barriers etc., it is still a disadvantage that
installation measures are required with the "cooperative target" in
the area of the machine to be safe guarded.
[0006] A device having two image recording units arranged at a
spacing from one another is disclosed by DE 197 09 799 A1. The
images, which are supplied to an evaluation unit from the two image
recording units, are subjected to a stereo image analysis in order
to obtain a three-dimensional image of the hazardous area. The
three-dimensional image is subsequently compared with a
three-dimensional reference model in order to detect foreign
objects, i.e. objects located inadmissibly in the hazardous area.
However, it has emerged from practical experiments with a
comparable arrangement that the reliability of detection with which
the foreign objects can be detected in the hazardous area is
inadequate. This applies at least to the extent that such a device
is to be used for protecting persons from automatically operating
machines. Extremely short reaction times (in the range of 30 ms)
and a virtually one hundred percent reliability of detection are
required for such an application.
[0007] A similar device and a corresponding method are also
disclosed by EP 1 061 487 A1, and what has been said above with
regard to the disadvantages also applies here.
[0008] DE 198 09 210 A1 discloses a device and a method for
monitoring what is called a scene. The scene is recorded with a
single image recording unit and compared with a reference image.
Individual points inside the scene are specifically illuminated by
means of a light source in order to obtain a marking or
individualization when recording the image. However, this method
has the disadvantage that the illumination of the scene, thus of
the hazardous area in the specific case of application, causes
susceptibility to faults owing to the incursion of scattered
extraneous light. Faults owing to extraneous light can occur, in
particular, in industrial production environments, for example with
welding robots or with large brake presses.
SUMMARY OF THE INVENTION
[0009] Against this background, it is an object of the present
invention to provide an improved method and a corresponding device
which allow to safeguard a hazardous area with-out a large outlay
on installation, but with adequate reliability.
[0010] According to one aspect of the invention, this object is
achieved by a method of the type mentioned at the outset, wherein
the scene analysis is carried out on the basis of the at least two
images using at least two algorithmically different methods, and
wherein the hazardous area is safeguarded guarded when at least one
of the analysis methods detects a foreign object.
[0011] According to another aspect, the object is achieved by a
device of the type mentioned at the outset, wherein the evaluation
unit is designed for carrying out the scene analysis using at least
two algorithmically different methods.
[0012] It is preferably suggested to carry out a three-dimensional
evaluation of the hazardous area on the basis of at least two
images that are recorded with mutually offset image recording
units. Appropriate scene analysis methods are known, for example,
from a book entitled "Three-Dimensional Computer Vision: A
Geometric Viewpoint" by Olivier Faugeras, MIT Press, Cambridge, or
from an article entitled "Fast Stereo Based Object Detection for
Stop & Go Traffic" by U. Franke/I. Kutzbach, INTELLIGENT
VEHICLES SYMPOSIUM, pages 339-344, Tokyo 1996, both documents are
incorporated herein by reference. An appropriate analysis method is
also disclosed by international patent application WO2004055732,
which is also incorporated herein by reference. On their own, those
methods known to date do not provide the reliability of detection
required to ensure that the hazardous area of an automatically
operating machine is reliably safeguarded. However, it has now
surprisingly emerged that the combination of different
three-dimensional scene analysis methods that are executed in
parallel with one another enables the reliability of detection
required for this application to be achieved.
[0013] Objects in the recorded images are detected by means of
separate scene analysis methods, and, preferably, their spatial
position is determined. The spatial position of the objects is
subsequently used to check whether the latter are located in the
monitored hazardous area, and whether "foreign objects" are
involved in this case. Additional external aids such as, for
example, a cooperative target or a specific marking of individual
points of the recorded hazardous area with light are not required.
The inventive solution therefore renders possible a low outlay on
installation. Moreover, a large degree of flexibility is rendered
possible by the novel device and the corresponding method because
of the numerous variants of modern evaluation systems.
[0014] The various scene analysis methods that are currently
available each have individual strengths and weaknesses, in
particular with regard to the reliability of detection and
susceptibility to faults. The respective weaknesses can be
compensated by combining different methods. The hazardous area is
safeguarded as soon as even only one of the scene analysis methods
detects a foreign object, the monitored machine thus being switched
off or otherwise brought to a state of rest, for example. The high
requirements that are placed on protective devices for machine
safety can thus be met surprisingly easy.
[0015] Moreover, the novel device has the advantage that diversity
is necessarily produced through the use of different methods for
scene analysis. The intrinsic fail-safety of the overall device
with reference to malfunctions and faults is thereby reduced. This
is of great significance with regard to the preferred field of
application, specifically safeguarding machines and
installations.
[0016] Stereo image processing is understood here to mean the use
of two cameras with approximately parallel optical axes and an
approximately identical focal length. A disparity map (3-D cloud of
points) is determined due to the fact that a pixel in the left-hand
image is assigned to a pixel in the right-hand image, this being
based in general on a small local environment of the respective
pixel.
[0017] By contrast, a three-dimensional scene analysis is
understood to be a method in which object parameters are adapted
such that there is an optimization of the parameters by means of a
target function that is established simultaneously by means of at
least two images of the object, as it is disclosed, for instance,
by Matusik et al., "Image-Based Visual Hulls", SIGGRAPH 2000 or by
international patent application WO2004055732, both of which are
also incorporated herein by reference. The reliability of detection
is raised compared to conventional stereo image processing methods,
since the recognition is performed here at a level of relatively
high information content.
[0018] In a refinement of the invention, the first and the second
images are subjected on the one hand to a correlation-based scene
analysis and, on the other hand, to a contour-based scene
analysis.
[0019] A correlation-based scene analysis in the meaning of the
invention is known, for example, from the article by
Franke/Kutzbach already mentioned above or from another article
entitled "Real-Time Stereo Vision for Urban Traffic Scene
Understanding" by U. Franke/A. Joos, IEEE INTELLIGENT VEHICLES
SYMPOSIUM, 2000, which is also incorporated herein by
reference.
[0020] A contour-based scene analysis is known, for example, from
the publication of Matusik, already mentioned above, or from above
mentioned international patent application. As has been shown in
practical experiments, the combination of these methods for scene
analysis delivers particularly good results with regard to the
reliability of detection and, furthermore, also with regard to the
intrinsic fail-safety of the overall device.
[0021] In a further refinement of the invention, at least one of
the methods processes only such image regions that differ
significantly from a reference image.
[0022] This measure has proved to be particularly advantageous for
the preferred field of application of the invention, since it is
possible thereby to raise the processing speed. Consequently, the
reaction times of the device can be shortened, and this renders
possible a high degree of flexibility in the fashioning of
protective zones, on the one hand, and a better protection of
operating personnel, on the other hand.
[0023] In a simple refinement of the method, only one image pair is
generated and the two scene analysis methods are carried out on it.
This refinement has the advantage that it can be realized simply in
terms of hardware. Under unfavorable conditions, however, there can
be the disadvantageous effect that a three-dimensional analysis is
not possible for structures in the image that are oriented along
the connecting line of the two image recording units.
[0024] In order also to be able to record such structures in
three-dimensions, it is advantageous to record at least one further
image of the scene with at least one further image recording unit.
These further image recording units should not lie on the same line
with the first two image recording units.
[0025] In order to keep the device simple in terms of hardware, it
is advantageous when three recorded images of the scene are
interpreted as two image pairs, the first and the second image pair
having an image in common in each case. This procedure can also be
extended correspondingly to the use of more than three images.
[0026] A particularly high degree of safety is achieved when a
number of scene analysis methods are applied to each image
pair.
[0027] In a particularly advantageous refinement, the inventive
device has at least a third image recording unit for recording a
third image of the hazardous area. The evaluation unit is set up in
this case such that at least one of the two scene analysis methods
is applied to the at least three images. The required computing
power and computing time are lowered by means of selecting a single
one of the two possible methods for scene analysis.
[0028] In a further refinement, a number of temporarily successive
groups of at least two synchronously recorded images are evaluated
in order to determine at least one movement parameter of a detected
foreign object, such as, in particular, speed, direction of
movement etc.
[0029] Due to this measure, it is possible to estimate more
accurately the degree of hazard originating from the monitored
machine or the like. An intelligent reaction to a detected foreign
object is possible as a result thereof. For example, in the case of
a foreign object that is moving only slowly inside the hazardous
area, it is possible firstly to generate an advance warning, while
the monitored machine is switched off directly in the case of a
quickly moving foreign object.
[0030] In a further refinement, a future exposure time of the image
recording units is set as a function of the current images.
[0031] The exposure time of the image recording units is matched
adaptively to changing environmental conditions by this measure.
The novel device can therefore react automatically to changing
environmental conditions, for example day-light influences. On the
one hand, the reliability of detection is raised thereby, and on
the other hand undesired false alarms can be better avoided.
[0032] In a further refinement of the invention, at least two of
the image recording units are arranged in a common equipment
housing.
[0033] This refinement has the advantage that a large part of the
required calibration work as well as the adjustment of the two
image recording units relative to one another can be undertaken by
the manufacturer at his works. The onsite outlay on installing the
inventive device is thereby simplified still further. Moreover, the
reliability of detection is raised even further for the entire
device by a precise calibration under laboratory conditions.
[0034] It goes without saying that the aforementioned features and
the features still to be explained below can be used not only in
the combination respectively specified, but also in other
combinations or on their own without departing from the scope of
the present invention.
BRIEF DESCRIPTION OF THE FIGURES
[0035] Exemplary embodiments of the invention are illustrated in
the drawing and explained in more detail in the following
description. In the drawing:
[0036] FIG. 1 shows a simplified block diagram of an inventive
device;
[0037] FIG. 2 shows a simplified flowchart for explaining the
inventive method;
[0038] FIG. 3 shows a sketch for explaining a preferred application
of the inventive device;
[0039] FIG. 4 shows a modification of the application shown in FIG.
3; and
[0040] FIG. 5 shows a flowchart of the inventive method in a
fashion adapted to the modification of the figure.
DESCRIPTION OF PREFERRED EMBODIMENTS
[0041] An embodiment of the inventive device is denoted in its
entirety in FIG. 1 by reference numeral 10. The device 10 includes
a camera module 12 and an evaluation unit 14. Illustrated inside
the camera module 12 are two image recording units 16, 18 (denoted
as "camera 1" and "camera 2" in the following). The image recording
units 16, 18 each can be complete, independently functional
cameras. As an alternative, however, individual parts of the two
cameras can also be used jointly such that there are essentially
two separate image sensors then, and, if appropriate, associated
optics in each case.
[0042] The image recording units 16, 18 are connected to an
evaluation unit 14, spatially somewhat set apart, via a bus 20. As
an alternative to this, it is also possible to integrate the two
image recording units 16, 18 and the evaluation unit 14 in a common
housing, or to use other kinds of an analog or digital interface.
The presently illustrated arrangement with a spatially separate
evaluation unit 14 has the advantages of a modular system, such as
the lower weight of the individual components, separate possibility
of replacement etc. Moreover, it is possible in this way for a
further camera module 22, illustrated here only schematically, to
be connected to the same evaluation unit 14 in a very simple
fashion.
[0043] Reference numeral 24 denotes a switching device that
contains relays or contactors, for example. The switching device
can be used to switch off the drives of a monitored machine,
installation or the like in a way known per se. The switching
device 24 can also be implemented, for example, in the form of a
failsafe PLC (programmable logic controller), i.e. what is called a
safety controller. It is preferred in this case when the evaluation
unit 14 is connected to the safety controller via a failsafe field
bus equipment, such as the SafetyBus.RTM. from Pilz GmbH & Co.,
Germany (one of the present assignees). The safety controller is a
PSS 3000 from Pilz GmbH & Co, for example.
[0044] The basic mode of operation of device 10 is described below
together with a preferred exemplary embodiment by means of FIGS. 2
and 3. Identical reference numerals respectively denote identical
elements in this case.
[0045] After the device 10 has been switched on, it firstly carries
out a self-test in accordance with step 30 in order to check
acceptable functioning. In the course of the self-test, the
individual components of the evaluation unit 14 and of the image
recording units 16, 18 are checked for acceptable functioning,
inter alia. In accordance with step 32, the outputs of the
switching device 24 are switched off, i.e. a monitored machine is
reliably brought into a defined, safe state.
[0046] In accordance with step 34, which is optional, a light
source is switched on as an additional illumination. This is done
particularly whenever the existing ambient brightness is inadequate
to carry out the subsequent image evaluation.
[0047] In accordance with steps 36, 38, the monitored hazardous
area is recorded by the image recording units 16, 18. In other
words, the image recording unit 16 supplies a first image of the
hazardous area, and the image recording unit 18 supplies a second
image thereof. In accordance with the exemplary embodiment
illustrated here, this is performed in parallel, i.e. with image
recording. units 16, 18 synchronized with one another.
[0048] What is called image preprocessing is subsequently performed
in accordance with step 42. Depending on the application, methods
known per se for conditioning the recorded images, for example an
FPN (Fixed Pattern Noise) correction, are carried out here.
Furthermore, the recorded first and second images are matched to
one another with regard to background brightnesses etc.
[0049] In accordance with steps 44, 46, the image pair consisting
of the first and the second image is subsequently evaluated by
means of two different scene analysis methods, in order to detect
foreign objects inside the hazardous area. In step 44, the first
and second images of the hazardous area are subjected to a
correlation-based image analysis. By contrast, a contour-based
image analysis is performed in step 46.
[0050] In a particularly preferred exemplary embodiment, the
contour-based image analysis includes a regional segmentation, i.e.
a division of the first and second images into individual regions
to be compared. It is then preferred to evaluate only regions that
differ significantly from a reference image.
[0051] Alternatively, however, in other exemplary embodiments it is
also possible to use an algorithm such as is described in
above-mentioned EP 1 061 487 A1 or in above-mentioned Matusik et
al., "Image-Based Visual Hulls", SIGGRAPH 2000, for example. For
correlation-based image analysis, it is preferred to use an
algorithm such as described in the articles by U. Franke cited
above. Basically, however, other scene analysis methods might also
come into consideration here.
[0052] In step 48, a decision is made on the basis of the results
of the two scene analyses as to whether a violation of the
protected space exists, i.e. whether an object has inadmissibly
penetrated into the monitored hazardous area. If this not the case,
the outputs of the switching device 24 are switched on in
accordance with step 50, i.e. the monitored machine is taken into
operation.
[0053] If a protective space violation by a foreign object has been
detected, the method branches to step 32, whereupon the outputs of
the switching device are switched off. (or remain so if a foreign
object was already detected in the previous method cycle). The
monitored machine etc. is thereby kept in its safe rest position.
It goes without saying that steps 34 to 50 are repeated recursively
in order to ensure that the hazardous area is safeguarded
continuously during operation of the device 10.
[0054] A preferred application of the novel device 10 is
illustrated in plan view in the upper part of FIG. 3, while the
associated side view is shown in the lower part.
[0055] The device 10 serves here to safeguard the hazardous area 58
which originates from an automatically operating robot 60. The
hazardous area 58 here lies inside a hemispherical enveloping
surface about the robot 60, and corresponds to its movement or
operating area. The camera module 12, which includes the two image
recording units 16, 18 in a common housing, is arranged above the
hazardous area and aligned with the latter.
[0056] Reference numerals 62 and 64 denote two staggered protective
zones, protective zone 62 forming a type of warning zone, while
protective zone 64 forms a true forbidden area. If a foreign object
66 enters the protective zone 62, this is detected by the device 10
on the basis of the scene analysis methods 44, 46. In one exemplary
embodiment, the evaluation unit 14 then switches the robot 60 into
a mode with reduced speed. An emergency shutdown of the robot 60
does not occur until violation of the protective zone 64.
[0057] In a further exemplary embodiment, upon violation of the
protective zone 62 the robot 60 is moved into a safe rest position
from where the robot 60 can undertake a restart as soon as the
device 10 detects that the protective zone 62 is no longer being
violated. If, by contrast, the protective zone 64 is being
violated, a manual restart is required.
[0058] The hazardous area 58 and the protective zones 62, 64 are
illustrated here in each case by enveloping curves in the shape of
a circle segment. In a preferred exemplary embodiment, the
hazardous area 58 and the protective zones 62, 64 are, however,
adaptively matched to the movements and operating states of the
robot 60. Here, the two scene analysis methods 44, 46 preferably
include additional algorithms by means of which parameters of the
foreign object 66, for example its shape, its direction and speed
of movement, can be identified and tracked. Dependent thereon, the
protective zones 62, 64 are adaptively defined about the robot 60
such that the foreign object 66 can be "close" to the robot 60
without being endangered. In this design, it is possible, for
example, to arrange a number of robots 60 close to one another in a
restricted space, collisions between them being prevented by means
of the device 10. Again, in a further exemplary embodiment,
workpieces (not illustrated here) that are being fed to the robot
60 can be distinguished from "unauthorized" foreign objects. The
mode of operation of the robot 60 can be optimized in this way
while the safety of operating staff participating is ensured at the
same time.
[0059] FIG. 4 shows a modification of the embodiment from FIG. 3.
Rather than a single camera module as in the case of FIG. 3, three
image recording units 16, 18, 22 are arranged here spaced apart
from one another in each case above the robot 60 and its protective
zones 62, 64. The image recording units 16, 18, 22 have extremely
wide-angled objectives so that they can cover the protective zones
62, 64 completely even when they are arranged only at a low height
above them.
[0060] In particular, the image recording units 16, 18, 22 can be
catadioptric cameras, i.e. cameras that use curved mirrors as
imaging elements, and have an annular field of view around a dead
angle lying on the extension of its optical axis. If these cameras
are aligned with the robot 60 by means of their optical axis, they
certainly cannot cover the robot themselves, but their field of
view overlaps well with the annular protective zones 62, 64 around
the hazardous area 58. Since the catadioptric cameras can thereby
detect any foreign object on its path into the hazardous area 58,
it is not disadvantageous that the hazardous area 58 itself largely
intersects their dead angles. Instead, this can signify a
simplification for the monitoring, since when it does not perceive
the robot the evaluation unit 14 also need not distinguish the
robot's movements from possible movements of a foreign object in
order to detect the latter.
[0061] The three cameras 16, 18, 22 are arranged such that the
connecting line between two cameras in each case lies in the field
of view of the third one. It is therefore always possible to detect
the position of a foreign object on such a connecting line by means
of images that are supplied by the third camera and one of the two
cameras lying on the connecting line.
[0062] The mode of operation of the apparatus from FIG. 4 is
described with reference to the flowchart of FIG. 5. The method
steps 30, 32, 34, 36, 38 are identical to those already described
with reference to FIG. 2, and are therefore not explained again. In
addition to the two image recording steps 36, 38 by camera 1 and
camera 2, a third image recording step 40 by camera 3 takes place.
There are thus three images present which have been recorded
simultaneously from different viewing angles and can be combined in
a different way to form image pairs. At least two different image
pairs are required, and the image pairs are combined in steps 70,
72 from the images recorded by cameras 1 and 2 or cameras 2 and 3.
A third image pair can be optionally combined from the images of
cameras 3 and 1 in step 74.
[0063] Two different scene analysis methods are applied to the at
least two image pairs thus obtained. In the simplest case, it is
possible, for example, to restrict oneself to carrying out a
contour analysis on the image pair of cameras 1 and 2 in step 76,
and to carrying out a correspondence analysis on the image pair of
cameras 2 and 3 in step 78. A higher measure of reliability of
detection is achieved when both analysis methods are carried out on
each of both image pairs (steps 80, 81), and when, if step 74 has
been carried out, the image pair thereof is also subjected to a
contour analysis 84 and a correspondence analysis 86.
[0064] Steps 48, 50 following the scene analysis are analogous to
those from FIG. 2, i.e. in step 48 it is decided by means of the
results of the preceding two, four or six scene analyses whether
there is a protective space violation. If this is the case, the
outputs of the switching device 24 are switched off, otherwise they
are switched on in step 50, i.e. the monitored machine is taken
into operation, or the outputs remain switched on if the relevant
machine is already in operation.
[0065] It will be appreciated that the contour analysis and the
correspondence analysis can be replaced by any other scene analysis
methods, or that it is also possible to carry out more than two
different scene analysis methods on a single image pair in order to
improve the reliability of detection of a protective space
violation. It is also possible to use methods of scene analysis
that do not undertake division of the images of the scene into
image pairs, but instead establish a three-dimensional description
of the scene by minimization from a target function to be
determined from all the images simultaneously. Such algorithms are
described, for example, in patent application DE 102 59 698-53,
published after the priority date, or in the above-cited
publication of Matusik et al. These two algorithms are preferably
to be applied in common to all the images in order to increase the
reliability of detection and to achieve a diverse system.
* * * * *